人工智涨ultra的行业观察

"FourCastNet requires 0.28 s for inferring a 24-hour forecast on a Tesla-A100 GPU (312 teraFLOPS), whereas Pangu-Weather needs 1.4 s on a Tesla-V100 GPU (120 teraFLOPS)."译文：FourCastNet 在 Tesla-A100 GPU 上推理 24 小时预报需要 0.28 秒，而 Pangu-Weather 在 Tesla-V100 GPU 上需要 1.4 秒。

"Each of the four deep networks was trained for 100 epochs, and each of them takes approximately 16 days on a cluster of 192 NVIDIA Tesla-V100 GPUs."译文：四个深度网络中的每一个都训练了 100 个 epoch，每一个在包含 192 块 NVIDIA Tesla-V100 GPU 的集群上大约需要 16 天。